Artificial Neural Networks, Adaptive and Classical Control for FTC of Linear Parameters Varying Systems

Abstract Three different schemes for Fault Tolerant Control (FTC) based on Adaptive Control in combination with Artificial Neural Networks (ANN), Robust Control and Linear Parameter Varying (LPV) systems are compared. These schemes include a Model Reference Adaptive Controller (MRAC), a MRAC with an ANN and a MRAC with an H ∞ Loop Shaping Controller for 4 operating points of an LPV system (MRAC-4OP-LPV, MRAC-NN4OP-LPV and MRAC-H ∞ 4OP-LPV, respectively). In order to compare the performance of these schemes, a coupled-tank system was used as testbed in which two different types of faults (abrupt and gradual) applied in sensor and actuators in different operating points were simulated. The simulation results showed that the use of ANN in combination with an adaptive controller for LPV-based system improves the FTC scheme, delivering a robust FTC system against abrupt and gradual sensor faults. For actuator faults, the only schemes that were fault tolerant were the MRAC-H ∞ 4OP-LPV and the MRAC-4OP-LPV (i.e. the MRAC-H ∞ 4OP-LPV was fault tolerant for actuator faults varying from 0 to 0.5 of magnitude).